Google AI weather outperforms conventional forecasting

Google’s AI-driven weather forecast model—called GraphCast—predicts weather faster, more accurately and up to 10 days in advance than other AI and conventional weather forecasting tools, reports the Nature journal.

Developed by DeepMind, GraphCast predicts weather phenomena in minutes and can run on a desktop computer rather than high-power-consuming supercomputers.

In September, it accurately predicted the landfall of Hurricane Lee three days in advance.

But the AI model can be faulty when predicting once-in-a-lifetime, record-breaking, extreme weather conditions because the large language models used to train it don’t have enough human-recorded data to feed it.

Nevertheless, the future of weather prediction has changed forever, and one can expect more AI-fed forecasts about their daily climate.